This dataset consists of time-series sensor data collected while performing hand gestures wearing a soft sensor-embedded glove. It includes both static and dynamic gestures, collected from 30 subjects to provide richer soft-sensor-based hand-gesture data.
SDGlove/
├── SDGlove.zip # Full dataset — 13,421 gesture records
│ └── s#/
│ └── {Gesture_num}-{Iteration_num}.csv
├── per_class_visualization.ipynb
├── per_class_s25.png # Per-class visualization example
├── Participant_info.md # Participant details
├── README.md
└── LICENSE
Research using hand gesture recognition based on this dataset can be applied in environments such as:
- Virtual safety training and robot teleoperation
- Healthcare
- Hand rehabilitation exercises
- Assistive communication devices
- Sign language education
- VR/AR devices
- User interface (UI) and menu navigation
- Virtual object and content interaction
- Data format: CSV
- Total number of data: 13,421 gestures
- Length per iteration: approximately 5 seconds
- Column description
Column Description sensor_1 Thumb MCP sensor_2 Thumb PIP sensor_3 Index MCP sensor_4 Index PIP sensor_5 Middle MCP sensor_6 Middle PIP sensor_7 Ring MCP sensor_8 Ring PIP sensor_9 Little MCP sensor_10 Little PIP - MCP: Metacarpophalangeal joints
- PIP: Proximal interphalangeal joints
-
Number of Participants: 30
-
Number of gesture classes: 12
- Static gesture (S1~S7)
- Dynamic gesture (D1~D5)
-
Data collection device: Mollison Hand (Feel the Same Inc., Korea, soft sensor-embedded glove)
- Sensor configuration: Soft sensors at each finger joint (measuring resistance values based on the degree of flexion at MCP and PIP)
- Sampling rate: Collected every 15ms
-
Collection procedure
- Glove worn on the participant's dominant hand
- Each gesture performed 40 times in random order.
- Gesture performance duration: 5 seconds
- Rest time between gestures: 3 seconds
-
Basic preprocessing (for incorrectly performed gestures)
- If a gesture from a different class was performed → relabeled to the corresponding class
- If a completely incorrect gesture was performed → removed
pip install numpy pandas matplotlib jupyterper_class_visualization.ipynb randomly selects one participant from SDGlove/, then randomly picks one instance per gesture class (S1–S7, D1–D5) and plots the 10 sensor channels (sensor_1–sensor_10) as time series, with all 12 classes shown together in a single figure (one subplot per class).
jupyter nbconvert --to notebook --execute --inplace per_class_visualization.ipynbExample: 12 gesture classes (S1–S7, D1–D5) for one randomly sampled participant (s25), each subplot overlaying the 10 sensor channels over time.
- If you want to cite our datasets, you can use our paper:
our bibtex
| Paper title | Journal/Conference | Year |
|---|
This project is licensed under the MIT License - see the LICENSE file for details.